Guerrilla Checkers: From Board Game to Machine Learning Environment.
June 3, 2025 Leave a comment
(photo: Brant Guillory)
“Guerrilla Checkers: From Board Game to Machine Learning Environment” is the title of a 2025 degree thesis by Niklas Krogerus at the Arcada University of Applied Sciences in Helsinki.
Abstract:
This is a software development work, in which the board game Guerrilla Checkers has been
implemented in Python and adapted for machine learning. Guerrilla Checkers is an asymmetrical board game for two by game designer Brian Train. It could be described as a combination of Checkers and Go. The project provides a new software implementation of Guerrilla Checkers and makes it available as a machine learning environment for the first time. The software is designed to be compatible with the most common Python library for machine learning environments, Gymnasium, as well as Petting Zoo, an extension of Gymnasium designed for training multiple machine learning agents simultaneously. While having ultimately failed to produce an agent capable of challenging a human opponent, the implementation is shown to have produced agents that perform significantly better than chance. The potential of achieving better results by refining machine learning techniques is indicated. The text also explores the basics of combinatorial game theory, including Ernst Zermelo’s foundational essay on chess and John Conway’s groundbreaking work On Numbers and Games, before making a rough mathematical assessment of how complex Guerrilla Checkers is.
The rest of it is in Swedish, so it’s beyond me… well, so is software development in general.
If you consider a set of game rules as a collection of algorithms that temporarily modify the behaviour of a human being, one would think that I would be a good programmer – but I’m not.
Still, thank you Niklas, for using my game as the basis for your work!
brtrain 
Recent Comments